Angela Strange

1. Key Themes

Ignorance as a Competitive Advantage in Emerging Markets

Santiago deliberately avoided being "too in the matrix" on Colombian financial services. His deep U.S. knowledge was a liability in the U.S. ("the Bank Holding Company Act of 1960 actually makes this unworkable") but his relative ignorance of Colombian constraints freed him to see the opportunity clearly.

"Had I been that into the matrix in Colombia, I would have been like, yeah, your company is never going to work out... But I didn't. So I think a little bit of ignorance is always bliss." [00:08:11]

Foundational Technology Bets That Compound Over a Decade

Two architectural decisions made seven years ago — a monorepo and event sourcing via Kafka — now underpin Addi's entire AI advantage. These were not obvious at the time; microservices was the dominant paradigm. The payoff only became visible once LLMs matured.

"We built a company on a monorepo as opposed to microservices. Monorepos today are all the rage. Anthropic's built on a monorepo. Google, obviously, famously is a monorepo company. But when we started the company, microservices was where it was at." [00:00:00]

"We built a company using an event sourcing architecture, which means that every single event that happens at this company gets logged... We generate over 10 million events per day." [00:21:20]

AI Deployment Starting From the Hardest Problem, Not the Easiest

Rather than beginning with customer service like most companies, Addi started AI deployment with legal response — a 48-hour constitutional deadline where failure means the CEO goes to jail. This forced the team to build robust data pipelines that then made every subsequent AI application easier.

"I remember when we were discussing this, I was like, but let's start with customer service like everyone else. And Carlos, my CTO, and then Mauro, they were like, no, this is the harder problem, but this is the one that scales. Because all the pipelines we need to build for the... by the way, we are saving a few lawyers, if not more on this." [00:24:30]

"Off the gate, 100% handling. And off the gate, it was 60% resolution. Because if you could resolve a lawsuit, you could resolve most customer service interactions." [00:25:29]

Single-Country Depth Over Multi-Market Breadth

Addi exited Brazil to go deep in Colombia, a move that looked like a retreat but was modeled on Kaspi's Kazakhstan playbook. The insight is that concentrated excellence in one market builds compounding moats — payments rails, merchant networks, consumer trust — that thin multi-market strategies cannot.

"He was like, look, this is a very unusual play. Equity investors will not understand this. Just ignore them for five years to 10 years. And then one day you'll come out and they'll be like, why didn't you call me? And you'll say, I tried many times." [00:15:14]

"We're in over a thousand cities, in a country with 1,100 cities, with our own payment rails. We do our own clearing, our own settlement, all in our own tech." [00:18:11]

Writing Everything Down as an Operational Superpower — Especially for AI

Addi's memo culture, borrowed from Amazon and driven by a desire to articulate "the why," turned out to be a massive AI accelerant. All SOPs being written down means agents can act on explicit context rather than tribal knowledge.

"We're also a remote company is a huge competitive advantage because it ensures all the context is explicit and it allows agents to work on that context... We have a lot of like formal APIs at the company that allow agents to grab them and run with them." [00:28:41]

"So would you mind just writing a note explaining why you think becoming a bank is a good idea? So it works. And by the way, in that process, we found a few things that we otherwise would not have found out." [00:42:21]

The North Star Metric as Organizational Survival Tool

Addi moved from heavy OKR cascades to a single North Star metric (profitability) that everyone in the company could name. Once achieved, they graduated it from a metric to a baseline expectation, then moved to 1-3 "L1 metrics."

"So it just focuses the discussion... now these days we even optimize our taxes because we're trying to optimize net income. So it just focuses the discussion." [00:38:57]

"The same way that no one says you should show up to work wearing appropriate clothing, that is profitability. It's just like an expectation." [00:40:07]

AI Is Causing Headcount to Run Structurally Below Budget

Addi is running 150 heads below budget while exceeding growth targets. This is not from cutting — they simply don't need to hire at the rate their growth historically would have required.

"We're running 150 heads below our budget while exceeding the growth... it doesn't mean that we are cutting heads. It just means that for the growth rate we had... we're running behind. So we cut that thing and we're going to keep cutting it again." [00:46:00]

Remote-First as an AI-Native Architecture Decision

Being remote-first forces all context to be explicit in writing, which is precisely what makes it machine-readable. This is a non-obvious structural advantage over in-person companies where context lives in hallway conversations.

"Being a remote company is a huge competitive advantage because it ensures all the context is explicit and it allows agents to work on that context." [00:28:41]

2. Contrarian Perspectives

Running a Colombian Company in English Is a Talent Advantage, Not a Tax

The conventional wisdom is that a local company should operate in the local language to reduce friction and attract local talent. Santiago found the opposite: English raises the company's perceived prestige and attracts ambitious local talent who want to work somewhere serious.

"Spanish speaking talent wants to work for great companies. And being in English, actually, there is a certain raising of the bar... it changes their mind. It's like we are here to do serious work because we're speaking in English." [00:27:12]

NPS as a Metric Is Worse Than Listening to Random Customer Service Calls

Kaspi's CEO told Santiago that for companies under a thousand employees, NPS is a terrible metric. Direct transcript review of random customer interactions is a more actionable signal.

"He's actually been public about this where he's like, I think NPS for a company with less than a thousand employees is a terrible metric. You should just listen to customer service calls. So, weekly business reviews, we always start with what happened to the customer last week. And we have two transcripts pulled at random." [00:13:51]

Start AI With the Hardest Use Case, Not the Easiest

Every company starts AI with customer service or internal search. Addi started with legal — 48-hour constitutional response deadlines under threat of CEO imprisonment. The pipelines required for that hard case made everything else trivial.

"Carlos, my CTO, and then Mauro, they were like, no, this is the harder problem, but this is the one that scales." [00:24:30]

Founders Should Build Their Own AI Stack Personally Before Mandating It Company-Wide

Santiago built his own EC2-based stack from scratch — doing his own DevOps, provisioning, API keys — before pushing AI adoption company-wide. Only after personally experiencing the full stack did he have the conviction to make it non-optional for the organization.

"I want to build my own. So I started from an empty EC2 instance. And from there, we built everything. And once I figured out, okay, if this is helpful for me at this scale, and if I'm able to do this kind of in between meetings, basically, then the entire company should be able to do this." [00:32:06]

Don't Let Ambition Follow Consensus — Consensus Plays Have No Alpha

The conventional LATAM playbook is Brazil first, then Mexico, then thin-layer expansion. Santiago argues that following conventional wisdom eliminates the alpha entirely, regardless of execution quality.

"Don't let your ambition fall prey of commercial wisdom. Because then there's no alpha, right? If you're a consensus play, there's just no alpha." [00:00:00]

3. Companies Identified

Addi

A buy-now-pay-later marketplace, payments platform, and now licensed bank in Colombia. Serves over 3 million consumers and 50,000 merchant partners across 1,000+ of Colombia's 1,100 cities, with proprietary payment rails, clearing, and settlement. Named third most innovative fintech globally by Fast Company.

"We serve over 3 million consumers, over 50,000 merchant partners... We're in over a thousand cities, in a country with 1,100 cities, with our own payment rails. We do our own clearing, our own settlement, all in our own tech." [00:00:00]

Kaspi

Kazakhstan's dominant fintech super-app, publicly listed in London. Referenced as the primary strategic model for Addi — proof that a single-country focus can produce a world-class fintech outcome.

"Everyone thinks you got to do everything at once... He was like, look, this is a very unusual play. Equity investors will not understand this. Just ignore them for five years to 10 years. And then one day you'll come out and they'll be like, why didn't you call me." [00:14:49]

Databricks

Data and AI platform. Addi partnered with Databricks five years ago to ingest Kafka event streams in real time and present them in vector form for LLMs and SQL form for ML models — the foundational data layer for all of Addi's AI.

"Five years ago now, we partnered with Databricks. So what that means is you, in real time, are able to ingest these events and showcase them in vector form for LLMs, SQL form for classical machine learning models." [00:22:13]

Anthropic

AI safety company and developer of Claude. Cited as a notable monorepo company alongside Google, validating Addi's early architectural decision.

"Anthropic's built on a monorepo. Google, obviously, famously is a monorepo company. But when we started the company, microservices was where it was at." [00:00:00]

TBC Uzbekistan

Fintech company in Uzbekistan run by former Tinkoff CEO Oliver Hughes. Mentioned as part of the Central Asian fintech ecosystem Santiago and his co-founder studied on their research trip.

"Oliver Hughes, who was the long-term Tinkoff CEO, now runs TBC Uzbekistan. And so we're like, let's go." [00:13:21]

Navia

Early probabilistic computing startup where Santiago was his first employee circa 2010. Doing what was effectively pre-transformer AI — programs that gave different outputs on each run. Described as two decades early.

"Navia was doing probabilistic computing back in 2010. And we used to say... our programs are so special because every time we run them, you get a different output. This is back in 2010." [00:03:50]

JPMorgan

Major U.S. bank. Santiago spent six years in Jamie Dimon's internal 10-person strategy group, which he describes as "a whole university in how to run companies and how to think about financial services."

"Jamie hated consultants. So he had banned all consultants, but he had a strategy group of 10 people... I was there for six years and it was incredible." [00:06:39]

Mercado Libre (MeLi)

Latin America's dominant e-commerce and fintech platform. Addi's new CFO came from an 11-year career there, and was most surprised by how well-organized Addi is relative to MeLi.

"The new CFO has an 11-year career at MeLi. And this board member asked him, hey, what is the thing that surprised you the most? And he's like, how well organized the company is." [00:05:18]

4. People Identified

Santiago Suárez

Founder and CEO of Addi. Came to the U.S. from Colombia on a full scholarship to Yale, was first employee at Navia, spent six years at JPMorgan's internal strategy group under Jamie Dimon, then founded Addi. Built one of Latin America's most technically sophisticated fintechs from Bogotá, now serving 25%+ of Colombia's population.

"We've been named the third most innovative company in fintech by Fast Company globally... in the last 24 months, a lot of our long-term investments have really paid off and have allowed us to ship extremely exciting product." [00:17:20]

Jamie Dimon

CEO of JPMorgan. Cited as one of the world's great operators. Santiago modeled his CEO education around learning from Dimon directly inside JPMorgan's elite internal strategy group.

"I wrote a list and we had Costco in it. I had Amazon in it. I had Wynn Casinos in it... leveraging McKinsey, Yale, everyone I knew, I just tried to get in a position to get a job somewhere close to these people. And the first one I came up with was this kind of legendary strategy group back at the time, which was 10 people at JPMorgan because Jamie hated consultants." [00:06:09]

Mikheil Lomtadze (Kaspi CEO)

CEO of Kaspi, Kazakhstan's fintech super-app. Santiago cold-messaged him four times on LinkedIn and flew to Kazakhstan to meet him. Gave Addi two foundational operating lessons: measure NPS by listening to customer service calls, and ignore equity investors for 5-10 years if you have a contrarian but correct strategy.

"After like three or four cold LinkedIn emails, he was like, okay, fine. I'm like, come. And then we go meet him. And he was so generous... I think NPS for a company with less than a thousand employees is a terrible metric. You should just listen to customer service calls." [00:13:21]

Oliver Hughes

Former long-term CEO of Tinkoff, now running TBC Uzbekistan. Mentioned as a world-class operator who moved to build fintech in an emerging Central Asian market.

"Oliver Hughes, who was the long-term Tinkoff CEO, now runs TBC Uzbekistan." [00:13:21]

Daniel (Addi Co-founder)

Santiago's co-founder at Addi. Referenced as a key strategic partner in annual research trips to study comparable markets, and in long-term vision discussions.

"What Daniel, my co-founder, and I try to do every year is we try to take a trip. To see things... But obviously, you know, the big prize is Kaspi." [00:12:56]

Sean McGuire

Investor in Square (later Block). Author of a notable Twitter/X post on how Elon Musk runs companies — specifically his habit of concentrating all attention on the single most important thing. This post directly influenced Addi's adoption of a North Star metric framework.

"I recently read this incredible post on Twitter by this Square investor Sean McGuire about... it was an incredible treatise on how Elon Musk runs companies. And we end up realizing he spends all his time on the most important thing. And very little time on everything else." [00:37:58]

General Partner at a16z, on Addi's board. Co-host of this episode.

Gabriel Vasquez

Partner at a16z focused on Latin America. Co-host of this episode.

5. Operating Insights

The CEO Must Build the Full AI Stack Personally Before Mandating Adoption

Santiago didn't just read about AI or watch demos. He provisioned his own EC2 instance, got his own API keys, did his own DevOps, and built an end-to-end stack himself in between meetings. Only then did he have the conviction and credibility to make AI adoption non-optional company-wide.

"I want to build the whole end-to-end stack. I don't want to just play because there's a lot of now agent wrappers... I started from an empty EC2 instance. And from there, we built everything. And once I figured out, okay, if this is helpful for me at this scale... then the entire company should be able to do this." [00:32:06]

Track Compile Time as a Proxy for Engineering Culture Seriousness

Addi tracked the time it took for code to compile after an engineer pushed to production as a major operational metric — when it crept from under 10 minutes to 33 minutes, they treated it as a crisis. This discipline is what makes a technology company actually a technology company.

"We would spend every week talking about how long it would take to compile when an engineer pushed a code into production. And it had gone from sub-10 minutes to 33 minutes. And we treated this as like, as a major problem... 100 engineers times 20 minutes twice a day, that's a lot of minutes." [00:34:30]

Never Settle Frivolous Lawsuits — Take Them to the Limit

Addi has a deliberate policy of never settling consumer protection lawsuits even when settlement costs less than litigation, because settling sets precedent that invites more suits. They have taken cases to the level just below the Supreme Court on principle.

"We've taken a couple all the way up to, like, the level below the Supreme Court, because we will just not settle. Because then people know." [00:23:38]

Hire Senior World-Class Leaders + Train Young Talent — Skip the Middle

Addi's talent model explicitly pairs world-class senior leaders (each search taking ~9 months) with young high-potential local talent, deliberately skipping mid-level hiring. This creates a high-ceiling development environment that attracts both tiers.

"We have a model where we bring world-class leaders and we become very good at hiring them, even though each search takes like nine months, true story. And then we train young up-and-coming talent. And that combo is extremely powerful." [00:29:11]

Once You Hit Your North Star Metric, Retire It to a Baseline Standard — Don't Keep Chasing It

After hitting profitability, Addi stopped discussing it in weekly reviews and instead declared it a non-negotiable floor — like showing up dressed appropriately. This freed cognitive bandwidth for new growth metrics while locking in the discipline.

"The same way that no one says you should show up to work wearing appropriate clothing, that is profitability. It's just like an expectation... Can we be unprofitable again? And we're like, no, you cannot be. It's just a necessary part of what we do." [00:40:07]

6. Overlooked Insights

Uzbekistan Is Producing Multiple Fintech Unicorns and Is Completely Under the Radar

Santiago mentioned in a single sentence, almost as an aside, that Uzbekistan has produced roughly three fintech unicorns — and that Oliver Hughes, the architect of Tinkoff's rise to one of Europe's most valuable banks, is now running TBC Uzbekistan. This is an extraordinarily strong signal. A market with Tinkoff-caliber leadership, low fintech penetration, and already multiple unicorn-scale outcomes is a pattern that historically precedes enormous value creation. Almost no Western investors are paying attention.

"Everyone knows about Kazakhstan, but also Uzbekistan. It's incredible. In Uzbekistan, there were like three fintech unicorns. Oliver Hughes, who was the long-term Tinkoff CEO, now runs TBC Uzbekistan." [00:12:56]

Addi Is Building Its Own Proprietary Transformer Model ("AdiDNA") on Its Own GPUs

In a single throwaway sentence, Santiago mentioned that Addi is training its own transformer model — not fine-tuning an existing one, but building from scratch on their own GPU infrastructure — using their proprietary closed-loop transaction and behavioral data. This is what Addi is calling AdiDNA. A fintech with 10 million daily events, seven years of transaction history, and a closed-loop payment network training its own foundation model is categorically different from a company using off-the-shelf LLMs. If this model works, it creates a moat in credit underwriting, fraud, and merchant intelligence that no external AI provider can replicate.

"We're investing in transformers. So we're building our own transformers using our own GPUs to train our own kind of like GPT of sorts. We call it AdiDNA with incredible early results." [00:32:34]